Aim. This paper is a report of the item selection process and evaluation of the content validity of the Critical-Care Pain Observation Tool for non-verbal critically ill adults. Background. Critically ill patients experience moderate to severe pain in the intensive care unit. While critical care clinicians strive to obtain the patient's self-report of pain, many factors compromise the patient's ability to communicate verbally. Pain assessment methods often need to match the communication capabilities of the patient. In non-verbal patients, observable behavioural and physiological indicators become important indices for pain assessment. Method. A mixed method study design was used for the development of the Critical-Care Pain Observation Tool in 2002-2003. More specifically, a four-step process was undertaken: (1) literature review, (2) review of 52 patients' medical files, (3) focus groups with 48 critical care nurses, and interviews with 12 physicians, and (4) evaluation of content validity with 17 clinicians using a self-administered questionnaire. Results. Item selection was derived from different sources of information which were convergent and complementary in their content. An initial version of the Critical-Care Pain Observation Tool was developed including both behavioural and physiological indicators. Because physiological indicators received more criticism than support, only the four behaviours with content validity indices >0·80 were included in the Critical-Care Pain Observation Tool: facial expression, body movements, muscle tension and compliance with the ventilator. Conclusion. Item selection and expert opinions are relevant aspects of tool development. While further evaluation is planned, the Critical-Care Pain Observation Tool appears as a useful instrument to assess pain in critically ill patients.